Introduction

Compelling evidence has shown that circulating branched chain amino acids (BCAAs) were consistently associated with type 2 diabetes (T2D), insulin resistance, obesity and other metabolic disorders [1,2,3]. A recent genome-wide association study (GWAS) performed in Caucasians identified a common variant rs1440581 near PPM1K (PP2C domain-containing protein phosphatase 1K) gene associated with higher ratio of circulating BCAAs to AAAs (aromatic amino acid) [4]. In addition, Mendelian randomization analysis in 16596 individuals indicated that the genetically predicted difference in leucine or valine by rs1440581 was causally related to T2D [5]. In an earlier study, we demonstrated that rs1440581 modified the effects of dietary intervention on long-term changes in weight loss and insulin sensitivity [6], and our findings were confirmed by an independent clinical trial [7].

Given the causal relation between BCAAs and diabetes observed in Caucasians, we assumed that such relation may persist in other populations such as Asians. Genetic marker could be better than biomarker in causality inference, because it is less likely to be affected by confounding and reverse causation. We used variant rs1440581 as a surrogate marker of circulating BCAAs, and tested the association between this BCAAs variant and risk of diabetes in a prospective cohort of Chinese.

In the present study, we investigated the relation of the BCAAs-associated variant with risk of incident T2D, and the longitudinal changes in glucose related metabolic traits. We further explored the potential gene–environment interactions in the T2D risk and changes in related metabolic traits. We selected rs1440581 as the genetic marker of BCAAs on considering that, it was found to be the strongest signal with circulating BCAAs and demonstrated previously having potential modifying effect on glucose metabolism and insulin resistance [6].

Materials and methods

Study population

The current prospective analyses were based on a Chinese cohort study designed to explore the effects of genetic, lifestyle and diet on the development of diabetes, obesity and early cardiovascular diseases. The study design has been described elsewhere [8, 9]. Briefly, the participants were recruited from the Songnan community in Baoshan district, Shanghai, China, in 2008 as baseline. First, the permanent residents aged 40 or older were invited to participate a glucose survey. During this phase, anthropometric measurements were performed and standard questionnaires were used to collect information on lifestyle factors, history of diseases and medications, etc. Also, fasting blood samples were collected for biochemical measurements. The study protocol was approved by the Institutional Review Board of Rui-Jin Hospital affiliated to Shanghai Jiao Tong University School of Medicine. Every participant provided written informed consent. Totally, 10185 subjects agreed and joined the first stage of investigation. Among them, 6919 participants provided DNA samples. In 2013, all the eligible participants were invited to have follow-up examinations to determine their glucose metabolism status. The follow-up investigation included a standard questionnaire to collect information on lifestyle factors, history of diseases and medications, etc. Anthropometric measurements, blood and urine sampling were performed. We excluded those participants without genotype information for rs1440581 variant (n = 19), those with T2D at baseline (n = 845) and those missing information to define glucose metabolism status at the follow-up examination (n = 12). Then, 6043 participants were included for the current analyses.

Data collection and biochemical measurements

Questionnaires and anthropometric measurements were performed at baseline and follow-up by trained investigators according to standardized protocols [9]. A questionnaire was used to collect information about lifestyles, such as tobacco smoking and alcohol drinking habits, history of diseases, surgery and medications. If the subject smoked cigarettes or consumed alcohol in the past 6 months on a daily basis, the current smoking or drinking status was defined as ‘yes’. The participants were divided into four groups according to their 5-year change status of smoking habits (categorical variable), as always-smokers, ex-smokers, never-smokers and new-smokers. Similarly, the change status of drinking habits was categorized into always-drinker, never-drinker, ex-drinker and new-drinker. The participants were asked whether their parents, siblings or children were ever diagnosed with T2D. If the answer was ‘yes’, we considered that they had a family history of diabetes. Body weight, height and waist circumferences were measured by trained physicians. Body mass index (BMI) was calculated as weight (in kilograms)/height (in square meters). Systolic and diastolic blood pressure (SBP and DBP) was measured on non-dominant arm three times using an automated electronic device (OMRON Model HEM-752, Omron Company, Dalian, China) with 1-min intervals after 10-min rest. The average value of the three measurements was used in analyses.

Venous blood samples were collected after a minimum 10-h overnight fast for biochemical analyses. Fasting plasma glucose (FPG) concentrations were measured using the glucose oxidase method through an autoanalyser (ADVIA-1650 Chemistry System, Bayer, Leverkusen, Germany). Fasting serum insulin (FSI) was measured using an electrochemiluminescence assay (Roche Diagnostics, Basel, Switzerland). Serum concentrations of triglycerides (TG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-c), high-density lipoprotein cholesterol (HDL-c) were measured by an autoanalyser (ADVIA-1650 Chemistry System, Bayer, Leverkusen, Germany).

SNP selection and genotyping

We selected rs1440581 as the genetic marker of BCAAs, which was the strongest signal for plasma BCAAs in GWAS [5] and demonstrated to have potential modifying effect on glucose metabolism and insulin resistance in our previous study [6].

DNA was extracted from the buffy coat fraction of centrifuged blood using commercial blood genomic DNA extraction kit (OSR-M102-T1, TIANGEN BIOTECH CO, LTD, Beijing, China) through an automated nucleic acid extraction instrument (OSE-M48, TIANGEN BIOTECH CO, LTD, Beijing, China) according to the manufacturer’s standard protocol. Specific assays were designed using the MassARRAY Assay Design software package (v3.1) (Agena Bioscience, http://agenabio.com/products/massarray-system/). Mass determination was carried out with the MALDI-TOF mass spectrometer and MassARRAY Type 4.0 software was used for data gathering (Sequenom, CapitalBio, Beijing, China) [9, 10]. The call rate for rs1440581 was 99.7% and the concordance rate was 100% based on genotyping 100 duplicates.

Incident cases ascertainment and definitions

T2D diagnosis evidence was included FPG level and glucose-lowering medication use. The known diabetes cases at baseline were excluded. Incident T2D was defined as according to the 1999 World Health Organization (WHO) criteria, as FPG ≥ 7.0 mmol/L or receiving anti-diabetic drugs or insulin injection at follow-up [9]. Incident impaired fasting glucose (IFG) was defined as 6.1 ≤ FPG < 7.0 mmol/L at follow-up, while < 6.1 mmol/L at baseline. The index of homeostasis model assessment of insulin resistance (HOMA-IR) was calculated as FSI (µIU/mL) × FPG (mmol/L)/22.5. The HOMA index of beta cell function (HOMA-β) was calculated as 20 × FSI (µIU/mL)/(FPG [mmol/L] − 3.5).

Statistical analyses

Statistical analyses were performed with SAS version 9.3 (SAS Institute, Cary, NC). Deviation of Hardy–Weinberg equilibrium for BCAAs-associated variant rs1440581 genotype was assessed by the Chi square test. Participants were categorized into three groups according to rs1440581 genotypes. ANOVA for continuous variables and the Chi square tests for categorical variables were applied for the comparison of difference according to genotypes at baseline. As with skewed distribution variables, such as serum TG, FSI, HOMA-IR and HOMA-β, data were logarithmically transformed before analyses.

Multivariate logistic regression models were used to evaluate associations of rs1440581 with incident T2D and IFG risk. Three models were used. Model 1 was unadjusted. Model 2 was adjusted for sex, baseline age and BMI, changes in BMI and smoking and drinking habits over the follow-up period. Model 3 was further adjusted for family history of diabetes, baseline serum lipids and HOMA-IR based on Model 2.

We further examined the potential interactions between rs1440581 and traditional risk factors on the risk of incident T2D and the 5-year change in the glucose-related metabolic traits, followed by stratified analysis. Multivariate logistic regression models were fixed by introducing a multiplicative term of rs1440581 and the stratified variable, rs1440581 and the stratified variable in the models simultaneously, after adjustments for sex, baseline age and BMI, and family history of diabetes.

A two-sided P value of less than 0.05 was considered statistically significant. We performed a Bonferroni’s correction to adjust for stratified analyses. A P value of 0.006 (0.05/8) was considered significant after adjustment of multiple comparisons.

Results

Baseline characteristics of the study participants and the changes in metabolic traits

The baseline characteristics of study participants according to rs1440581 genotypes were presented in Table 1. The mean age was 57.8 ± 9.3 years and 37.4% were males. The major allele (T) frequency of PPM1K variant rs1440581 was 50.1% and the genotype distribution was in Hardy–Weinberg equilibrium (P = 0.81). No significant difference was found in age, sex, BMI, blood pressure, FPG, FSI, HOMA-IR, or HOMA-β between the PPM1K rs1440581 genotypes.

Table 1 Baseline characteristics of the study participants according to BCAAs-associated variant rs1440581
Full size table

The mean value of BMI over 5 years was increased by 0.61 (SD, 2.20) kg/m2. The FPG and HOMA-IR levels were significantly increased; while the FSI and HOMA-β levels decreased (Supplemental Table 1).

Associations of BCAAs related variant rs1440581 with incident T2D and IFG

During the 5-year follow-up, 576 (9.5%) non-diabetic participants at baseline developed T2D, and 695 participants (12.0%) with normal glucose regulation developed IFG. The incidence of T2D was increased along with C-alleles, either in non-diabetic participants (P for trend = 0.01, Fig. 1a) or in those normal glucose regulation (P for trend = 0.0005, Fig. 1b). However, no significant increasing trend for incidence of IFG with C-allele was found (Fig. 1c).

Fig. 1
figure 1

a The incidence of T2D in non-diabetic participants at baseline by rs1440581 genotypes. b The incidence of T2D in participants with normal glucose regulation by rs1440581 genotypes. c The incidence of IFG in participants with normal glucose regulation by rs1440581 genotypes. d The incidence of IFG + T2D in participants with normal glucose regulation by rs1440581 genotypes. P for trend was calculated using Cochran–Armitage trend tests

Full size image

Logistic regression analyses showed that each C-allele in rs1440581 was associated with a 20% increased risk of diabetes (odds ratio [OR] = 1.20, 95% confidence interval [CI] [1.05–1.36]), after adjustments for sex, baseline age and BMI, changes in BMI and smoking and drinking habits over the follow-up period, family history of diabetes, serum lipids and HOMA-IR (Table 2). In participants with normal glucose regulation at baseline, rs1440581 C-allele was significantly associated with risk of incident T2D (OR = 1.29, 95% CI [1.11–1.50]) and composite outcome of combined T2D and IFG (OR = 1.15, 95% CI [1.04–1.27]), but not IFG (P = 0.64) (Table 2).

Table 2 Association of BCAAs-associated variant rs1440581 with incident T2D and IFG
Full size table

Associations of 5-year changes in BMI with changes in glucose-related metabolic traits according to rs1440581 genotypes

We observed that BMI increase was positively associated with 5-year increase in FPG, Log10_FSI, Log10_HOMA-IR and Log10_HOMA-β (all P < 0.0001); whereas we did not find significant associations between BCAAs related variant rs1440581 and changes in these glucose-related metabolic traits (Supplemental Table 2). We further performed the gene–environment interaction analysis and found that rs1440581 significantly modified the effects of weight gain on increase in Log10_FSI and Log10_HOMA-IR (both P for interaction = 0.003), after adjustments for sex, baseline age, BMI, change in smoking and drinking habits over the follow-up period, and baseline values for the corresponding outcomes (Table 3). In the C-allele carriers, the association of BMI increases with increase in Log10_FSI (β ± SE, 0.027 ± 0.002) and Log10_HOMA-IR (0.030 ± 0.003) were more prominent than that in the T-allele carriers (0.019 ± 0.002 and 0.020 ± 0.002, respectively). We did not detect a significant interaction between BMI change and the genetic variant on changes in FPG and Log10_HOMA-β (both P ≥ 0.16, Table 3).

Table 3 Stratified analysis of associations of the 5-year changes in BMI with changes in glucose-related metabolic traits according to rs1440581 genotypes
Full size table

Interactions between genetic variant and traditional risk factors

We found a nominal significant interaction of rs1440581 variant with baseline SBP and DBP levels on risk of incident T2D, after adjustments for the confounders (P for interaction = 0.01 and 0.02, respectively). The C-allele was associated with a stronger risk of T2D with SBP ≥ 120 (OR = 1.28, 95% CI [1.11–1.47]) and DBP ≥ 80 mmHg (OR = 1.36, 95% CI [1.15–1.60]), as compared to the lower blood pressure (the corresponding OR = 0.85 and 0.99). No significant interactions were observed between the genetic variant and sex, baseline age, BMI, smoking and drinking status, and FPG (Fig. 2).

Fig. 2
figure 2

Stratified analysis of BCAAs-associated variant rs1440581 on the risk of incident T2D. P values were calculated using logistic regression model after adjustments for sex, baseline age and BMI, and family history of diabetes. P for interaction was tested by logistic regression analysis, by introducing a multiplicative term of rs1440581 and the categorized stratified factor, rs1440581 and the categorized stratified factor in the model simultaneously, after same adjustments as above

Full size image

Discussion

In this prospective investigation performed in a large size of Chinese cohort, we reported that the most strong and leading genetic signal of blood BCAAs levels from the GWASs, rs1440581 near PPM1K gene, was significantly associated with incident T2D risk. Furthermore, rs1440581 significantly modified effects of BMI changes on changes in IR, as indicated in FSI and HOMA-IR level. The association of BMI increase with increases in FSI and HOMA-IR was more prominent in C-allele carriers than that in T-allele.

A limited number of studies have examined effects of genetic determinants of serum BCAAs levels on the risk of incident T2D and longitudinal changes in glucose-related traits; and particularly the potential interactions with traditional risk factors in prospective cohort studies. Previous studies demonstrated a positive association of circulating BCAAs and the risk of T2D in different ethnicity groups [11,12,13]. A recent systematic review and meta-analysis reported BCAAs (leucine, isoleucine, and valine) prospectively associated with pre-diabetes and/or incident T2D using comprehensive high-throughput metabolomics techniques [14]. In a very recent Mendelian randomization analysis performed in a combined large sample size of Caucasians, a genetically predicted difference of one standard deviation in amino acid level was associated with an odds ratio of T2D of 1.44 for isoleucine, 1.85 for leucine, and 1.54 for valine [5]. In our present study, we demonstrated a BCAAs related genetic variant rs1440581 was significantly associated with higher risk of incident T2D. The results were also in line with our previous findings that rs1440581 affected changes in insulin sensitivity. Our data provided supporting evidence for a causal relationship between serum BCAAs and an increased risk of T2D. According to the Mendel’s Second Law, the random assortment of alleles/genotypes transferred from parent to offspring at the time of gamete formation [15, 16]. The independent distribution of alleles/genotypes allows a study relating a health outcome to a genetic variant would not be affected by confounders that often distort the interpretation of findings. Besides, because the random assignment to genotype occurs at conception, the association between a genetic variant and a disease is less affected by reverse causation [17].

Furthermore, we found the rs1440581 variation significantly modified effects of BMI increase on increase in FSI and HOMA-IR level during 5 years of follow-up. These findings were consistent with and confirmed what we have identified in the previous study [6]. In the gene–diet interaction analysis, we reported that individuals carrying the C allele of the BCAAs-associated variant rs1440581 might benefit less in weight loss and improvement of insulin sensitivity than those without this allele when undertaking an energy-restricted high-fat diet [7]. These findings suggested the rs1440581 might play an important role in modifying weight regulation and related metabolic benefits.

PPM1K was identified as a T2D gene by a systems genetics approach and rs1440581 is at upstream of this gene [18]. The expression of PPM1K influenced insulin secretion in human islets. The PPM1K is a kind of branched chain α-ketoacid (BCKA) dehydrogenase phosphatase and acts importantly in the metabolism of BCAAs [19, 20]. A recent study in mice demonstrated that the expression of PPM1K in muscle biopsies during an oral glucose challenge failed to increase in patients with T2D, suggesting that impaired α-keto acid dehydrogenase (BCKD) activation might partly mediated the association between IR and higher BCAA levels [21]. Menni et al. reported that BCKA 3-methyl-2-oxovalerate was the strongest predictor of IFG and the genetic variant rs1440581 had the strongest associations with all BCAAs and all BCKAs in their study [22]. Further studies are needed to determine whether the rs1440581 polymorphism affects PPM1K expression, and further influences the risk of developing diabetes.

The strengths of this study included the large well-defined community-based prospective cohort study design and a relatively large number of participants. Moreover, it is the first time to assess a GWAS-identified and validated BCAAs-associated genetic variant rs1440581 with risk of incident T2D and its effect on longitudinal changes in glucose-related metabolic traits. Several limitations should also be acknowledged. First, the 75-g oral glucose tolerance test was not performed at baseline examinations. Absence of the 2 h post-loading glucose might misclassify participants with T2D into non-diabetic group at baseline and underestimate new cases during the follow-up period. Second, we did not measure circulating BCAAs levels in the participants, which prevented the potential analysis of the association between the genetic variant and circulating BCAAs levels. Although, a genetic marker could be a surrogate for a biomarker in revealing causal relationship according to Mendelian randomization theory [16, 23], we acknowledged that the present study is a genetic association analysis, not a standard Mendelian randomization analysis. Testing circulating levels of BCAAs and AAAs would have provided more evidence and strengthened the conclusion. Third, we studied a single one genetic variant associated with BCAAs levels. Rs1440581 near PPM1K gene was first identified in large-scale GWAS and consistently validated as strongest and leading signal of genetic predisposition to diabetes risk in relation to circulating BCAAs [4, 5]. Whereas, a genetic risk score consisting more established loci related circulating BCAAs levels as the instrument variable are warranted. Lastly, our population is limited to middle-aged and elderly Chinese subjects. Further studies from other demographic groups or ethnicities are expected.

In conclusion, we found that the BCAAs related variant rs1440581 was significantly associated with the risk of incident T2D in Chinese. This genetic variant might modify the longitudinal changes in IR in relation to body weight changes. It would provide supportive evidence for high-risk stratification of population and personalized prevention to prevent diabetes. Replication of the findings in other populations and potential mechanism exploration are warranted.